With the increasing usage of video in different applications, creation and management of video databases have received considerable attention [Oomoto and Tanaka, Chua and Ruan, Hacid et. al.]. To support associative retrieval and maintenance of video data, it is necessary to develop a suitable video data model which also enables the semantics of the video segments to be formally dealt with. In video data modeling and database literature temporal and object- oriented approaches have often been advocated. A temporal approach focuses on temporal order of video segments. An object-oriented approach, on the other hand, supports representations of objects and their spatial and topological relations in a video frame. It is also possible to combine both these schemes to examine the temporal variation of object properties and their spatial and topological relations in a video segment. Hacid et. al. have suggested an extension of Datalog-based logic formalities to define the semantics of video segments.However, the existing models do not give adequate importance to the behaviour of objects present in the video. This is specially important when a video is used as a tool to analyze the dynamic behaviour of a system. For example, an echocardiogram video, which is a very important tool for diagnosis and monitoring of cardiac diseases, depicts the dynamics of the heart chambers and valves. The segments of an echocardiogram should therefore be more intimately related to the dynamics of the heart, such as systolic and diastolic phases of heart. Again, Magnetic Resonance (MR) images are also used to study the condition of the heart. In medical diagnosis, it will be immensely useful if parallel composition of echocardiogram and MR images are shown in different windows with proper synchronization. The semantics of synchronization should be guided by the behavioural semantics of the underlying objects. Thus, a cardiologist can look into the video segments corresponding to systolic or diastolic phase of the cardiac cycle of a patient as obtained from the echocardiogram and MR recordings, in two different windows on the screen. In this paper, we introduce a video data model where the semantics of a video is specified by the properties (static and dynamic) of the objects present in the video. The structure and behaviour of the objects are represented following UML paradigm. The object dynamics is captured by its states and state transition. Thus, for an echocardiogram database, the primary objects are heart, its chambers and valves. The heart can be in two states, namely, systolic and diastolic. The proposed model associates a guard condition with a video condition. A guard condition is a well-formed formula of first-order predicate calculus involving object attributes, states and state transitions. The paper describes the semantics of a video algebra involving projection and composition operators.